Intergenerational Occupational Mobility in a Rural Economy

M. Shahe Emran1

George Washington Universityand IPD, Columbia University

Forhad ShilpiDECRG, World Bank

This Version: June 2007

ABSTRACT

This paper presents evidence on intergenerational occupational mobility from agricul-ture to the nonfarm sector using survey data from Nepal. In the absence of credible instru-ments, the degree of selection on observables is used as a guide to the degree of selectionon unobservables a la Altonji et. al. (2005) to address the unobserved genetic correlationsrelevant for occupation choice. The results show that a moderate ability correlation caneasily explain away the observed partial correlation in non-farm participation between thefather and a son. In contrast, the partial correlation in occupation choice between motherand daughter is much stronger, and is unlikely to be driven by genetic correlations alone.The results suggest that mother’s nonfarm participation plays a causal role in daughter’schoice of nonfarm occupation, possibly because of “cultural inheritance” through role modeland learning effects, and transfer of reputation and social capital.

Keywords: Intergenerational Occupational Correlations, Non-Farm Participation, Gen-der effect, Cultural Inheritance, Selection on Observables, Selection on Unobservables

JEL Codes: J62, O12

1We are grateful to Todd Elder and Chris Taber for very useful comments on an earlier draft, especiallyregarding the econometric issues. We would also like to thank Paul Carrillo, Bob Goldfarb, Don Par-sons, Dilip Mookherjee, Bhashkar Mazumder, David Ribar, and Stephen Smith for useful comments, andTodd Elder for sharing STATA and MATLAB routines for implementation of the econometric estimation.The standard disclaimers apply. Emails for correspondence: shahe.emran@gmail.com (M. Shahe Emran),fshilpi@worldbank.org.

(1) Introduction

The evolution of income distribution, inequality and occupational structure across gen-

erations has attracted increasing attention in recent economic literature2. This renewed

interest reflects a widely shared view that strong intergenerational linkages in socioeco-

nomic status may reflect inequality of opportunities and thus have profound implications

for poverty, inequality and (im)mobility in a society. A large body of econometric studies

focusing mainly on developed countries finds that intergenerational correlations in earn-

ings are positive and statistically significant, ranging between 0.14 to 0.50 (see Blanden

et. al. (2005) and Solon (1999, 2002)). There is a (relatively) small empirical literature

in economics, again mostly in the context of developed countries, that indicates significant

positive correlations between parents and their children in occupational choices (see, for

example, Lentz and Laband (1983), and Dunn and Holtz-Eakin (2000) on U.S and Sjogren

(2000) on Sweden). Economic analysis of intergenerational mobility in developing coun-

tries, however, remains a relatively unexplored terrain3, even though the importance of

such analysis has been duly recognized in the recent literature.4 In this paper, we present

evidence on the intergenerational occupational correlations in the non-farm sector in a de-

veloping country, Nepal.5 Although there is a substantial literature on the determinants of

non-farm participation (see Lanjouw and Feder (2001) for a recent survey), the issue of in-

2See, for example, Arrow et al. (2000)., Dearden et. al. (1997), Mulligan (1999), Solon (1999, 2002),Birdsall and Graham (1999), Fields (2001), Fields et. al. (2005), Bowles et. al. (2005), Blanden et.al. (2005), WDR (2005), Mazumder (2005), Hertz (2005), Mookherjee and Ray (2006), Bjorklund et. al.(2006).

3This is exemplified by the fact that Solon (2002) refers to only two studies on developing countries in hissurvey of economic mobility (Lillard and Kilburn (1995) on Malaysia, and Hertz (2001) on South Africa).The recent analysis of economic mobility in the context of developing countries include Lam and Schoeni(1993), Behrman et. al. (2001), Fields et. al. (2005), Dunn (2004). There is, however, a substantialsociological literature that analyzes occupational mobility in both developed and developing countries (see,Ganzeboom et. al. (1991), Morgan, (2005)).

4For example, Bardhan (2005) identifies intergenerational economic mobility as one of the importantbut under-researched areas in development economics.

5As emphasized by Goldberger (1989), there are some important advantages to focusing on occupationalmobility rather than income mobility. For example, the intergenerational linkages might be stronger foroccupation choice (relative to income), and focusing on income correlations “could lead an economist tounderstate the influence of family background on inequality” (P.513).

1

tergenerational linkages has so far not received any attention. Understanding occupational

mobility in a rural economy is, however, important for poverty alleviation, as mobility from

agriculture to nonfarm is often an avenue to escape poverty trap (WDR (2005); Lanjouw

and Feder (2001)). In the presence of strong intergenerational linkage in occupational

choice, the standard cost-benefit analysis is likely to under-estimate the long-run social

returns to policy interventions that encourage non-farm participation, as the intergenera-

tional multiplier effect is ignored. Non-farm participation often leads to ‘visible’ income

contribution by women and thus positively affects their bargaining power.6 A different

but powerful argument derives from the role of non-farm entrepreneurship in the long-run

structural transformation of an economy. A dynamic non-farm sector can be the seedbed

for experimentation and development of an entrepreneurial class that eventually graduates

to industrial activities, as was the case in Japan’s rise to a modern industrial state from

late Tokugawa to Meiji period (See Smith (1988)).

Our focus in this paper is on two issues: (i) intergenerational occupational persis-

tence beyond the observed determinants like education and assets, and the unobserved

genetic correlations across generations, and (ii) gender effects in occupational mobility.

The literature on the intergenerational economic mobility has been fraught with economet-

ric challenges that arise from the unobservability of the genetic characteristics (ability and

preference transmissions across generations), and the partial correlations observed in the

data (from multivariate regressions) might be driven largely by such unobserved genetic

correlations between parents and children.7 The distinction between genetic transmissions

and other sources of intergenerational linkages can be important from policy perspective.

If the observed intergenerational occupational correlation is primarily due to ability cor-

relations, then the role public policy can play to influence economic mobility is relatively

6Women’s work in agriculture is usually unpaid and remains invisible in developing countries.7Genetic transmissions relevant for occupational choice include both ability and preference (especially

the degree of risk aversion). However, the focus of the literature has been on ability correlations. In whatfollows, we couch the discussion primarily in terms of ability correlations, following the literature.

2

limited.8 In contrast, when other tangible (like education and wealth) or intangible (like

role model effects9, learning and reputation externalities and social capital) environmental

factors are important in intergenerational linkages, it provides additional arguments for

policy interventions.

In the absence of experimental data10, the standard approach to identification when

facing unobserved heterogeneity like ability correlations is to look for credible instrumental

variables (IV). In the specific context of occupational mobility, the econometric challenge

is to find exogenous variation that affects parental occupation choice but does not have any

independent effect on children’s occupation choice. However, most of the potential can-

didates for IV such as family background variables that affect parent’s occupation choice

tend to affect children’s choice also. Thus it is difficult to defend the exclusion restriction.

Moreover, the common practice of using parental characteristics (like parental education)

as IVs is also suspect, as they are likely to be correlated with the unobserved common

ability, and thus likely to violate the exogeneity criterion. There is a small literature in

economics that uses adoption as a quasi-experimental design to isolate the effects of envi-

ronmental factors in intergenerational economic mobility (see, for example, Bjorklund et.

al. (2006); Plug (2004); Plug and Vijverberg (2003), Scaerdote (2002)). A third strategy

is to use twin samples to try to isolate the effects of nurture from that of nature (see, for

example, Behrman and Rosenzweig (2002)). However, these studies using adoption or

twin samples are confined mostly in the developed countries where such data of reliable

quality are available. In the absence of quasi-experimental data on adoptions and twins

8We, however, note that one should not take the distinction between genetic transmissions and otherenvironmental factors too far. The evidence from Behavioral Genetics shows that there may be significantdynamic interactions between nature and nurture in determining human behavior (see, for example, Plominet . al. (2001)).

9The definition of role model adopted so far in economic literature is not uniform. While Durlauf (2000)defines role model as the influence of “characteristics of older members” on the “preferences of youngermembers”, Manski (1993) and Streufert (2000) define it as observations on older members whose choicesreveal information relevant for the choice of younger members. In this paper, we adopt a broad view thataccommodates both of these definitions.

10Designing and implementing a randomized experiment that can generate the data required for under-standing the intergenerational occupational persistence can be challenging on both ethical and feasibilitygrounds.

3

or any credible identifying instruments, we exploit the econometric methodology recently

developed by Altonji, Elder and Taber (2005, 2000) (henceforth AET (2005, 2000)) which

provides a way to gauge the importance of common (across generations) unobserved hetero-

geneity in explaining an observed partial correlation and thus helps to determine if at least

part of the observed partial correlation is causal (due to environmental factors). We note

that genetic transmissions (ability and preference) influence both parents’ and children’s

nonfarm participation decisions and hence can be treated as an unobserved (and correlated)

determinant of nonfarm employment choices of both generations. This allows us to utilize

a battery of recently developed econometric tests to ascertain whether the observed inter-

generational occupational correlations can be attributed solely to the unobserved ability

correlations between children and their parents. 11

The results from the econometric analysis are as follows. The univariate probit estima-

tion indicates the presence of strong and positive intergenerational occupational correlations

along gender lines (mother-daughter and father-son) even after controlling for a rich set of

regressors including education (own, parents’ and spouse’s), assets (inherited land), age and

ethnicity (caste/tribe). The estimated occupational linkages from univariate probit can,

however, at least in principle, be due entirely to genetic transmissions across generations.

The evidence from the econometric analysis using the AET (2005) methodology shows that

this might actually be the case for the observed occupational correlation between the father

and son; even a moderate correlation of unobserved ability can explain away the estimated

occupational linkage completely. The intergenerational occupational correlation between

mother and daughter, on the other hand, is much stronger and unlikely to be driven solely

by unobserved ability correlations. The evidence thus suggests that at least part of the

correlation between mother and daughter is likely to be causal due to “cultural inheritance”

(Bjorklund, Jantti, and Solon (2007)) that includes role model effects, learning externalities

11Note that although the unobserved ability correlations have been the focus of much of the recentliterature on mobility (and returns to education), the unobserved common factors in our context includeother common determinants of occupation choice like social norms.

4

and transfer of reputation and social capital from parents to children, among other things.

The substantive conclusions above are very robust, confirmed by alternative economet-

ric techniques as developed by AET (2005. 2000): (i) sensitivity analysis using a bivariate

probit model, and (ii) estimates of lower bounds on intergenerational occupational corre-

lations.12 In case of daughters, the lower bound estimate of intergenerational occupational

correlation with mother is 0.66 with an implied marginal effect of 0.14 and t-value of 4.8.

The 95 percent confidence interval for the marginal effect is [0.07 0.24], which does not

include zero. These results suggest that the genetic correlations account for about half of

the partial correlation between the mother and a daughter given that the marginal effect

in the univariate probit model is 0.30. The other half of the intergenerational correlation

can be attributed to cultural inheritance by a daughter from her mother in the form of role

model effects, learning externalities and transfer of reputation and social capital, among

other things.13 In the case of sons, the lower bound estimate is negative and statistically

insignificant which implies that the observed partial correlation may be driven entirely

by the unobserved factors common to both generations. The results from the sensitivity

analysis yield the same conclusions as above.14

The rest of the paper is organized as follows. Section 2 provides a conceptual framework

that underpins the empirical work presented in the subsequent sections. The section 3

12The empirical methodology proposed by AET (2005, 2000) can be used to provide a lower bound onintergenerational occupational correlation under the assumption that the ‘selection on observables’ is atleast as large as the ‘selection on unobservables’. As we discuss in more detail later in the text, theassumption that the selection on observables dominates the selection on unobservables is a natural onein the context of intergenerational occupational mobility analyzed in this paper. The univariate probitmodel assumes “no selection on unobservables” and thus can be thought of as the upper bound estimateof the intergenerational correlation.

13Note that the environmental factors like role model effects and learning externalities only affect the oc-cupational choice of children and thus are NOT subsumed under the common intergenerational correlation.

14Following AET (2005, 2000) we also estimate the bias in the partial correlation estimates from uni-variate probit. The estimates of the bias might be useful as robustness check as they are not dependent ondistributional assumptions. However, as noted by AET (2005, 2000), the bias estimates are based on thestrong assumption that the bias in the linear projection is similar to the bias in the probit equation. Sincethis assumption is difficult to justify, we chose not to present the bias estimates (available upon request).We, however, note that the bias estimates also lend strong support to the central conclusions discussedabove.

5

discusses the data and variables, while the next section presents some preliminary evidence.

Section 4, arranged in a number of subsections, presents the main empirical results that

focus on gauging the role played by unobserved common determinants of occupational

choice across generations following the approach due to AET (2005, 2000). Section 5

concludes the paper with a summary of the main findings.

(2) The Conceptual Framework

In this section, we outline a simple model of participation in nonfarm sector highlighting

different channels through which intergenerational linkages may operate, especially the role

cultural inheritance might play through factors like role model and learning effects and

transfer of reputation and social capital from parents to children. The model is based on

the standard occupational choice model but is augmented to capture the essentials of the

intergenerational linkages.15

There are two sectors in the economy: agriculture (A) and non-farm sector (N). At

the beginning of the working life, every person in the economy decides which sector to

work for. Each individual is endowed with an innate ability θi ∈ [0, 1] that captures the

attributes that are relevant for non-farm sector. So the higher is θi the better suited an

individual is for non-farm employment. A fundamental source of intergenerational linkage

arises from the fact that the genetic endowments of a child ( θi) are likely to be correlated

with those of parents. The innate ability parameter θi is not known with certainty and

every individual has to form an estimate utilizing all the available information contained

in an appropriately defined information set.

In addition to ability, every individual is endowed with a vector of initial capital stock

ki comprised of human, financial and physical, and social capital. The higher is the level

of ki the higher is the probability of getting a better paid non-farm job. Parents can

influence this initial capital stock ki through their investment in a child’s human capital

(e.g. education) and their transfer of financial and physical capital. In addition, parental

15The model utilized here can be viewed as an extension of the celebrated contributions of Becker andTome (1979 and 1986) and the recent extensions proposed in Sjogren (2000).

6

occupation can also influence their offsprings’ human capital as children can gain valuable

skills and experience by observing their parents at work, and by informal apprenticeship in

parents’ work place, especially when the nature of occupation is such that the workplace

is in close proximity to home.16 The parents, when successful in self-employment, often

transfer significant reputation capital and a rich social network (social capital) to their

children. An individual’s earning from an occupation, Yi, is determined by innate ability (

θi) and endowment of capital stock (ki).

At the beginning of the working life, individual i takes the endowment of capital and the

estimate of ability (ki, θi) as given, and optimally chooses the occupation di ∈ A,N. Let

the information set available to individual i choosing occupation is denoted as Ωi which

include ki and θi. Let F (Yi | A; Ωi) denote the conditional distribution of income (Yi)

when individual chooses agriculture and the information set is Ωi.The associated probability

density function is denoted as P (Yi | A; Ωi). The preference of an individual i is represented

by a concave utility function, Ui(.), that reflects, among other things, the risk preference.

17

We define the expected utility from choosing agriculture as:

Vi(A, Ωi) ≡∫

Ui(Yi)P (Yi | A; Ωi)dYi

Analogously the expected utility from choosing non-farm sector is:

Vi(N, Ωi) ≡∫

Ui(Yi)P (Yi | N ; Ωi)dYi

16As noted by Lentz and Laband (1983), this proximity of work place to home is an important factorbehind the observed strong intergenerational following in occupations like agriculture. This proximity isalso important in case of the household based activities common in the microcredit programs.

17The preferences of a child are likely to be correlated with those of her parents. In addition, parentscan also induce changes in children’s preferences by acting as their role models (Durlauf, 2000). Theintergenerational correlation in preferences implies, for example, that, on an average, the children of theparents more inclined to taking risk will themselves be risk takers, and thus are more likely to becomenon-farm entrepreneurs.

7

The individual chooses non-farm employment iff the following holds:18

Vi(N, Ωi)− Vi(A, Ωi) ≥ 0 (1)

The probability that an arbitrary individual i drawn from the population will decide to

work in the non-farm sector is Pr(Vi(N, Ωi)−Vi(A, Ωi) ≥ 0). At the heart of the occupation

selection process is the formation of expectation about payoffs from different options using

the information set Ωi. A critical element of the information set is the occupational choices

of the parents as they reveal two types of relevant information: (i) information about ones

own genetic endowment (or innate ability), (ii) information about the characteristics of

a certain occupation. For example, if parents are successful (unsuccessful) non-farm en-

trepreneurs, the estimate of children’s ability to be successful in similar occupation will be

revised upward (downward). Another important channel is that revelation of information

might reduce the uncertainty about the parental occupation, and thus induce risk-averse

children to prefer the parental occupation to other alternatives. Thus, the information

revealed by parental choices (and their outcomes) can influence children’s occupation de-

cision through their effects on the conditional distribution function of income Yi giving

rise to role model effects (Manski 1993; Streufert, 2000). For example, consider a child’s

participation decision in non-farm sector (di = N). The parental role model effects imply

that the conditional distribution of income when parents are in non-farm F (Yi | N ; Np, Ωi)

is stochastically dominant over the conditional distribution of income with neither of the

parents is in non-farm F (Yi | N ; Ap, Ωi).19

The model presented above can also be used to explain intergenerational correlations

running along gender line. First, the genetic transmissions might have a gender dimension.

For example, the preference of a daughter (son) is likely to be more aligned with that of her

(his) mother (father) compared to that of her (his) father (mother). Second, and probably

18Assuming that the tie is broken in favor of non-farm sector.19Note that given a concave utility function both first and second order stochastic dominance are suffi-

cient.

8

the most important factor behind gender effects in intergenerational linkages in occupa-

tional choices, is the gender dimension in role model effects. The information revealed by

the choices (and consequent outcomes) of an older member of a society will be more infor-

mative for the choices of a given younger member the closer he/she is to the younger person

in an appropriately defined socioeconomic space. The individuals can be grouped together

by partitioning the socioeconomic space according to different exogenous (like ethnicity,

gender) or endogenous (like schooling) characteristics. The finer the partitioning the more

informative is the information revealed by the choices of a member of a given group for

the other members of that same group. It follows that, given the membership in a family,

gender creates a finer partitioning, and the mother becomes the natural role model for the

daughter, and the father for the son. This has also implications for learning by doing and

observing as the daughter (son) ‘sees’ and ‘hears’ primarily what her (his) mother (father)

does and says. Another potential channel for gender effects is that the effects of parental

social capital might run predominantly along gender lines; the mother’s social network

might be more easily accessible to a daughter. Moreover, social norms regarding gender

roles might also contribute to gender effects in occupation choice. The existence of gender

effects for a daughter means that the conditional distribution of income from non-farm

employment when mother is in non-farm F (Yi | N ; Nm, Ωi) stochastically dominates the

conditional distribution with father in non-farm F (Yi | N ; N f , Ωi).

For the econometric estimation, we can now employ a standard probit model taking

inequality (1) as the basis for our empirical specification. Specifically, we consider the

binary response model (with slight abuse of notation):

Ni = 1 N∗i ≡ Vi(N, Ωi)− Vi(A, Ωi) ≥ 0 , (2)

For estimation we impose linearity and assume that the latent variable N∗i is generated

from a model of the form

9

N∗i = αpN

pi + X ′

iγ + εi (3)

Where Xi ⊆ Ωi is a vector of explanatory variables and εi is the idiosyncratic random

disturbance term. In the econometric analysis, the vector of explanatory variables Xi is

required to include regressors that can control for heterogeneity across individuals in terms

of preferences (Ui), and the productivity, and pay-off information contained in Ωi. Equation

(2) forms the basis of much of our empirical analysis. A complete list of explanatory

variables Xi is provided in appendix table A.1.

(3) The Data

The data for our analysis come from the Nepal Living Standard Survey (NLSS) 1995/96.

The NLSS consists of a nationally representative sample of 274 primary sampling unit

(PSUs) selected with probability proportionate to population size, covering 73 of the 75

districts in Nepal. In each of the PSUs, 12 households were also selected randomly (16

households in the Mountain regions) providing a total sample size of 3373 households. With

an average household size of about 5.6, the survey collected detail information for 18855

individuals. Of the total individual level sample of 18347 individuals for whom parents

can be identified from the data, nearly 71 percent reported participation in the labor force,

but about 20 percent did not report any occupation.20 For the rest of the sample (9417

observations), 10 percent are either child labor (less than 14 years of age, 9 percent) or

too old (more than 70 years of age) and thus are dropped.21 But some of the parents did

not either participate in the labor force or report their labor force participation, further

reducing the size of the sample.22 Moreover, a number of the PSUs showed no employment

20A large fraction of those not reporting occupation are in fact child labor with age 14 years or less.21Note that the empirical results reported in the following sections remain unchanged even if we use any

other cut-off age (e.g. dropping those below 20 years of age and above 65 years of age and so on).22For 8394 individuals left in the sample, both parents reported participation in the labor force in the case

of 6874 individuals, and for rest of the observations, employment status of either father (347 observations)or mother (1173) are missing. Note that if we use dummies to capture the missing parental information,the sample size can be increased but the qualitative results remain unaffected.

10

diversification which are dropped to avoid perfect fits in the regression analysis. Splitting

the sample between males and females, we have a final sample of 2037 observations (in 152

PSUs) for daughters and 2919 observations (in 242 PSUs) for sons.

The NLSS contains detail information on employment by sectors and by occupations at

individual levels. The survey is unique in the sense that it contained an entire section of

questionnaire on parental information, including level of education, sector of employment

and place of birth. From the occupation information, we define our dependent variable as

a binary variable taking the value of one if an individual is employed in nonfarm activities

and zero otherwise.23 Similarly we define separate indicator variables for mother and father

showing their employment in nonfarm sector.

Table 1 reports the basic statistics on employment status of daughters and sons. The

(unconditional) probability of being employed in nonfarm sector is estimated to be around

44 percent for a man and 16 percent for a woman. In our data set, average participation

rates of father and mother are around 20 percent and 8 percent respectively. A comparison

of sons and daughters’ employment status conditional on father’s and mother’s employment

status reveals that the probability of being employed in non-agriculture sector is markedly

higher for both sons and daughters if father or mother were employed in non-agriculture as

well. We also tested the significance of difference between probabilities of being employed in

nonfarm sector by parent’s employment status (farm vs nonfarm). The test results reported

in Table 1 indicate that in all cases, the null hypothesis of no difference can be rejected with

a P-value equal to 0.00. According to Table 1, mother’s participation in non-farm sector

appears to have a larger effect, compared with father’s non-farm participation, on both

sons’ and daughters’ probability of participation in non-farm sector. The intergenerational

occupational linkage appears to be much stronger for daughters relative to sons.

23Non-farm is defined as non-agricultural, i.e., excludes SIC one digit code ‘0’. Non-farm thus includesrural industries and services.

11

(4) Preliminary Evidence

With some indication of positive intergenerational correlations between parents’ and

children’s occupational choices, we turn to more formal regression analysis. Starting from

a simple Probit regression of son’s and daughter’s occupations on parental occupations,

we take a sequential approach in presenting the results, introducing an array of control

variables in subsequent steps. The upper panel in Table 2 reports the regression results for

daughters and lower panel for sons.

Column (1) in Table 2 reports the coefficients of N f (father in non-farm) and Nm

(mother in non-farm) in the regression for son’s and daughter’s participation in nonfarm

sector. The results from the probit regression without any controls show that mother’s

non-farm participation has a significant positive influence on daughter’s probability of par-

ticipation in the same sector. The marginal effect of a mother’s participation in nonfarm

sector (Nm) is estimated to be 0.43 which is large compared to the daughter’s average

probability of participation in non-agriculture of 0.16. In contrast, father’s participation

in nonfarm (N f ) appears to have no statistically significant effect on daughter’s likelihood

of being employed in the same sector. The results for sons reported in the lower panel of

Table 2 indicate significant positive correlation between father and son’s employment in

the nonfarm sector. The marginal effect of father’s employment in nonfarm sector (N f ) is

around 0.15 which is statistically significant at 1 percent significance level. Compared with

father, mother’s nonfarm participation (Nm) has a smaller marginal effect (0.10) which is

significant at 5 percent level.

The next set of results reported in column (2) of Table 2 includes a large number of

household and individual level characteristics as control variables. The access to non-farm

jobs may depend on the personal networks that often run along ethnic group/caste (see,

for example, Dreze, Lanjouw and Sharma, 1998). To capture the variations in access to

non-farm jobs, we include a set of dummies depicting the ethnicity (caste and tribe) of the

individual in the regression. We also include dummies showing if there is any short/long-

12

term migrant in the household, as migration frequently occurs on the basis of personal

networks. A set of household variables including household size and composition are also

added to the set of control variables. As discussed in the conceptual framework, human

and financial capital variables are important links in the intergenerational transmissions

of socioeconomic status. In addition to the level of education, we include the age of an

individual as a human capital variable representing the work experience.24 The education

levels of parents and spouse are also included as additional human capital controls. The

inherited land (as the most important form of collateral), remittances received, and travel

time to the nearest commercial bank are included as controls for access to capital25. We

include an individual’s marital status to account for taste and/or life-cycle related hetero-

geneity. The summary statistics for these explanatory variables are presented in appendix

Table A.1. A comparison of daughters’ and sons’ samples in Table A.1 shows that the

difference in the means of individual and family characteristics across parental occupation

is much more pronounced in case of sons sample. This indicates that selection might be

relatively more important for the sons’ sample.26

The results reported in column (2) in Table (2) show that the added set of regressors

are powerful determinants of nonfarm participation decision. With the inclusion of these

controls, the Pseudo R2 increases from 0.10 to 0.52 in daughter’s sample and from 0.03 to

0.23 in son’s sample. However, the addition of the powerful set of controls does not affect

the estimated intergenerational partial correlations in any significant way. The marginal

effect of Nm (mother in non-farm) is estimated to be 0.41 which is virtually identical to

the estimate from the regression with no controls (0.43). It is still highly statistically

significant (t = 8.04). In the sons sample, the marginal effect of father’s employment in

nonfarm sector (N f ) is estimated to be 0.16 which is also nearly identical to our earlier

estimate from regression with no controls (0.15).

24In addition, age and its squared term capture any cohort effect.25The ethnicity dummy may also capture access to credit.26This implies that in our case the girls’ sample is more like the preferred C8 (catholic school 8th graders)

sample in AET (2005).

13

Although the regressions in column (2) include a large set of individual and household

level controls, the intergenerational correlation in occupation may still result, spuriously,

from the fact that parents and children may face similar labor market opportunities. For

instance, if both parents and children live in an area with better non-farm opportunities

(say location of a textile mill), then intergenerational correlation in non-farm participation

may be an artefact of not adequately controlling for non-farm opportunities in the regres-

sion. To control for unobserved location specific heterogeneity in non-farm opportunities,

we included village level fixed effects in the estimation (151 and 241 village dummies in

daughter’s and son’s regressions respectively). The village fixed effects may also capture

other village specific determinants of occupational choice like peer effects and agglomera-

tion forces. In addition, we define the share of non-farm employment in total employment

of an individual’s age cohort in her district of birth as an additional control for labor market

opportunity. This may capture the time varying part of labor market opportunities in a

village.

The results from regressions with village fixed effects are reported in column (3) of

Table 2. The addition of village level fixed effects as well as a measure of intertemporal

labor market opportunity leads to an increase in the explanatory power of the regressions

further. The Pseudo R2 of the regression is 0.62 in daughter’s sample and 0.53 in son’s

sample. Despite the inclusion of such a large number of controls (village plus household plus

individual level controls), the qualitative results regarding intergenerational occupational

correlations remain largely unchanged. Although the marginal effect of Nm (mother in

non-farm) on a daughter’s nonfarm participation declines, it is still large (0.30) and is

statistically highly significant with a t-value=6.33. The marginal effect of N f (father in

non-farm) on a son’s nonfarm participation is now 0.10 and is statistically significant at 1

percent level. Consistent with the available evidence in the literature on income mobility

(see, for example, Solon, 2002), the evidence also indicates that the cross gender effects are

not important as they are not statistically significant in column (3) of Table 2.

14

(5) Genetic Transmissions and Intergenerational Correlations

The results discussed so far show that the intergenerational occupational correlations

between parents and children run along gender lines (father-son and mother-daughter).

The evidence indicates that the estimated partial correlations are not solely due to the

‘tangible’ determinants of occupational choice like education and assets, as they are already

controlled for in the regression. The results can not be driven by village level factors

like peer effects and geographic agglomeration as we include fixed effect. However, as

discussed before, an important question from the policy perspective is how much of the

partial correlations uncovered in column 3 of table 2 is causal due to environmental factors

like role model effects and learning externalities as opposed to pure genetic correlations in

occupational choice.

In the absence of credible identifying instruments, we utilize a number of ways to ascer-

tain whether the observed intergenerational correlations can be explained solely in terms

of unobserved ability correlations (and other unobserved common determinants). They

include (i) sensitivity analysis a la Rosenbaum and Rubin (1983), Rosenbaum(1995) and

AET (2005, 2000); (ii) estimation of lower bounds on intergenerational correlations using

the technique developed by AET (2005, 2000).

(4.1) Sensitivity Analysis

The regression results presented in the previous section (Table 2) demonstrate that the

inclusion of a large and powerful set of controls does not lead to a substantial weakening

of intergenerational occupational correlations especially for the daughter. This suggests

that the selection on observables is dominant and a relatively small amount of selection is

due to unobservables. We now explore the question whether a small amount of selection

on unobservables can explain away the estimated partial correlations in intergenerational

occupational choices in table 2. Consider the following bivariate probit model for individual

i.

15

Ni = 1(αpNpi + Xiγ1 + δjωj + ξ > 0), (4)

Npi = 1(Xiβ1 + δjωj + u > 0) (5)

u

ξ

∼ N

0

0

,

1

ρ

ρ

1

(6)

where Ni (also Npi ) is a binary occupation choice variable which takes the value 1 for

non-farm and zero otherwise, ωj is the village dummy (fixed effect) included to control

for unobserved and observed community level determinants including labor market oppor-

tunities and peer effects. We estimate the magnitudes of intergenerational correlations

for different values of the correlation (ρ) between the unobserved determinants of nonfarm

participation of parents (u) and children (ξ).27 The vector of explanatory variables (X) is

the same as that in the regression results presented in column (3) of Table 2. However,

the inclusion of village level fixed effect in the regression describing parental participation

in nonfarm sector (Np) causes problem in estimation as there are cases of perfect fit due

to the absence of parental occupational diversification in a village (all 0s or 1s for the

occupation dummy). When we exclude such cases, the sample size reduces to 1126 in

daughter’s sample and 2547 in son’s sample. The results for these restricted samples are

reported in columns 2 and 4 of Table 3 for daughters and sons respectively. An alterna-

tive approach that keeps the sample size same as in Table 2 relies on an index of village

fixed effects estimated from the simple probit regressions reported in column 3 of Table

2. The results for these unrestricted samples are presented in columns 1 and 3 of Table

3 for daughter’s and sons respectively. Following AET (2005), the sensitivity analysis is

performed for ρ = 0.1, 0.2, 0.3, 0.4, 0.5. Note that the correlation coefficient ρ represents

27As discussed in AET (2005, 2000) the bivariate probit model above is identified because of nonlinearity.However, such identification based on functional form alone in the absence of valid instruments is treatedwith skepticism in applied literature (termed “weak identification”). In what follows, the bivariate probitmodel is treated as underidentified and thus the sensitivity analysis is performed across alternative valuesof ρ.

16

only that part of genetic correlation across generations which influence the occupational

choice.

For daughters, the results from the unrestricted sample show that the marginal effect

of the mother’s employment in nonfarm sector declines to 0.22 when ρ =0.10, and to 0.15

when ρ =0.20. The estimated marginal effect continues to decline with an increase in ρ

but is still positive though small in magnitude when ρ is as high as 0.50. Interestingly, all

the values of marginal effect are also statistically significant at 5 percent or less except for

the case when ρ =0.50. The conclusions derived from the restricted sample reported in

column 2 are similar to that of the unrestricted sample; the marginal effects are, however,

in general, larger in magnitude. These results suggest that barring sampling error, the

unobserved genetic correlations pertinent to occupation choice would have to be greater

than 0.50 to explain away the entire effect of Nm (mother in nonfarm) on a daughter’s

nonfarm participation.

In the case of sons, the marginal effect of father’s nonfarm participation becomes numer-

ically small (0.05) and statistically insignificant in the unrestricted sample when ρ = 0.10.

For values of ρ equal to or greater than 0.2, the marginal effect becomes negative. The

results are again very similar in the case of the restricted sample. These results suggest

that the estimated effect of N f (father in nonfarm) on son’s nonfarm participation may be

entirely driven by common unobserved factors like genetic transmissions.

(4.2) Lower Bounds on Intergenerational Occupational Linkages

The sensitivity analysis above indicates that the value of ρ would have to be larger

than 0.50 to completely explain the effect of mother’s nonfarm employment on that of

daughters found in Table 2. But there is no estimate of ρ in the literature which we can

use as a benchmark. The available evidence from Behavioral Genetics shows that both the

genetic transmissions and environmental factors are important in the correlation between

the parents and children, especially for complex traits and behavior.28 The problems in

28For example, the correlation between IQ scores of parents and children is around 0.5 (Plomin et. al.,

17

pinning down a plausible range for ρ are more daunting in our case as other explanatory

variables like education (both children’s and parents’), ethnicity (i.e., caste and tribe),

and assets are likely to pick up a substantial part of this correlation.29 In the absence

of any plausible way of judging the magnitude of the genetic correlations relevant for

occupation choices in a rural economy as captured by ρ, we utilize an approach suggested

by AET (2005). This allows us to estimate both the magnitude of ρ and bounds for the

intergenerational correlations.

To illustrate the basic insights behind AET (2005) approach, we consider equation

(3) (with village fixed effects added). It defines the latent variable N∗i that determines

children’s participation in nonfarm sector as:

N∗i = αpN

pi + X ′

iγ1 + δjωj + ε (7)

where Npi is the dummy variable for nonfarm participation by parents and ωj is the

village fixed effect for village j where individual i lives in. Let Np∗ is the latent variable

such that Np = 1 if Np∗ > 0 and zero otherwise. We can define the linear projection of

Np∗ on X ′γ, ω and ε as (for notational simplicity the subscript is dropped) :

Proj (Np∗|X ′γ1, ω, ε) = φ0 + φX′γ1X′γ1 + ω

′δ + φεε (8)

Following AET (2005), we can interpret φX′γ1 as the “selection on observables” and

φε the “selection on unobservable”. However, unlike AET (2005), we use a village level

fixed effect to sweep off the observed and unobserved village level determinants. This

implies that the selection on observables (φX′γ1) and unobservables (φε) both represent

only the individual characteristics. An advantage of this formulation is that it fits well

with the notion that the ‘unobservables’ are like ‘observables’. An alternative approach is

2001, Griffiths et. al. 1999). This correlation includes both the effects of nature (heritability) and nurture(familiality).

29This implies that the value of ρ relevant for our analysis should be smaller than otherwise.

18

to include the village fixed effects as part of the observables. The argument is that the

location of an individual is an observable characteristic.30 The linear projection of Np∗ in

this case becomes:

Proj (Np∗|Z ′γ2, ε) = φ0 + φZ′γ2Z′γ2 + φεε ; Z = (X,ω) and γ2 = (γ1, δ) (9)

The advantage of this formulation is that it is more likely to satisfy the condition

that selection on observables is dominant which helps in deriving the lower bound on

intergenerational occupational linkage (see below).31 We perform the analysis under these

alternative interpretations (equations (8) and (9)).32 Note that in the case of univariate

probit regressions, the maintained assumption is that there is no selection on unobservable,

i. e, φε = 0.

AET (2005, 2000) and Altonji, Conley, Elder, and Taber (2005) (henceforth ACET

(2005)) show that selection on observables can be used as a guide to selection on unob-

servables. They point out that in many applied economic applications, it is a natural

assumption that the selection on observables dominates the selection on unobservables

which leads to the following conditions in our case (analogous to condition (3) in AET

(2005)):

φX′γ1 ≥ φε ≥ 0 (10)

φZ′γ2 ≥ φε ≥ 0 (11)

Following AET (2005), we can implement the econometric estimation under the above

restriction(s) and treat the estimate of αp (equation (7)) corresponding to the case of

30We thank Chris Taber for pointing out the alternative interpretations of the fixed effects.31Since location choice is endogeneous, the village fixed effects will capture some of the unobserved

individual characteristics which are common to the villagers.32A third alternative is to exclude the fixed effects altogether and use village level observed controls

(share of non-farm). The conclusions of this paper remain unchanged in this formulation, although thelower bound estimates are larger than reported here.

19

equality of selection on observables and unobservables (i.e., for example, φX′γ1 = φε) as

the lower bound on the part of intergenerational occupational linkage that is not driven

by genetic transmissions. The inequality conditions (10) and (11) above are eminently

plausible in our case due to the following considerations.33 First, as pointed out earlier,

the addition of a set of rich and powerful determinants of occupation choice affects the

strength of intergenerational linkages only marginally although the Pseudo R2 goes up

dramatically. For example, in daughter’s sample, the Pseudo R2 increases from 0.10 to

0.52 when we include a rich set of determinants of occupational choice including education

levels of children, parents and spouse, inherited land, and ethnicity. The estimated partial

correlation in non-farm participation by mother and daughter is, however, barely affected

(it declines from 0.43 to 0.41). This indicates that (i) the observables explain a large part

of the variations in non-farm participation, and thus leave room for only a limited role for

the unobserved individual characteristics; (ii) the estimated partial correlation is robust to

possible inclusion of additional controls (if such data were available). Second, the data for

our analysis come from a multipurpose household survey which was conducted primarily

for poverty assessment. Since the role of non-farm occupations as an avenue to escape

poverty traps in a low income agrarian economy is much discussed (Lanjouw and Feder,

2001), it is only natural that the survey includes rich information on the determinants of

non-farm participation identified in the recent literature. This means that these observable

characteristics are likely to pick up a substantial part of the unobserved genetic correlations

relevant for occupation choice, a point mentioned earlier, but worth emphasizing again

here. This also means that the selection on unobservable genetic endowment captured

in φε will be much smaller in our analysis. Third, we can decompose the error term in

the occupation choice by children as in equation (4): ξ = ξ1 + ξ2 where ξ1 is the part of

selection on unobservables that is common to both generations but is determined at the time

of parental occupation choice, and ξ2 represents the unobserved shocks that occur during

33We are grateful to Chris Taber and Todd Elder for clarifying the relevance of the conditions (10) and(11) in our analysis.

20

the children’s occupation choice. As shown by AET (2005), this implies that selection on

observables is greater providing additional justifications for inequality conditions (10) and

(11) above.34

In the case of bivariate probit (equations 4-6), the lower bound estimate of αp can be

estimated by imposing the following conditions depending on the treatment of fixed effect:

ρ =Cov(Z ′β2, Z

′γ2)

V ar(Z ′γ2); β2 = (β1, δ) (12)

ρ =Cov(X ′β1, X

′γ1)

V ar(X ′γ1)(13)

Table 4 reports the estimates of the lower bounds on the intergenerational partial cor-

relation (i.e., lower bound estimates of αm and αf ). The inclusion of village dummies leads

to computational difficulties and convergence problems in the estimation of the bivariate

probit model. The results discussed earlier in Table 3 show that the index of village fixed

effects estimated from univariate probit model performs equally well as village level dum-

mies in controlling for spatial labor market opportunities. To avoid the non-convergence

34We note that three assumptions need to be satisfied for the point identification within the AETapproach. Two of the assumptions ensure that the selection on observables is equal to selection onunobservables (AET (2000), P.4). They are: (i) the observables are a random subset of a large set ofpotential determinants, (ii) the number of explanatory variables is large, none of the elements dominatingthe distribution of Np

i or Ni. The detailed regression results presented in appendix Table A.2 show that thesecond assumption is approximately satisfied in our context. As discussed above in the text, in our case,the assumption of random selection of observables does not hold, as the survey includes rich informationon the determinants of non-farm participation given its focus on poverty assessment. Thus the selection onobservables is likely to be larger than the selection on unobservables. This implies that when estimation isdone under the restriction of equality of selection on observables and unobservables, the estimated causaleffect of parental occupation will underestimate the true magnitude of the intergenerational linkage and thuscan be treated as the lower bound. The third assumption below when coupled with the equality of selectionon observables and unobservables can provide point identification. The third assumption essentially saysthat the observables and unobservables are nearly uncorrelated. This assumption (a stricter version of it)is invoked in most of the empirical applications in economics with its focus on the endogeneity of a singlevariable (like the extensive literature on returns to education) and thus assumes that all other controlsare not correlated with the error term. This, however, is not a natural assumption and may be difficultto justify in many applications. As emphasized by AET (2000), even if this assumption is not satisfiedexactly, the Monte Carlo evidence indicates that the resulting bias is small. Also, AET (2005) and ACET(2005) note that for actual implementation of the approach what is needed is that the relevant assumptionsare satisfied approximately.

21

problems, we use this index in the regressions reported in Table 4. The first panel reports

the results from bivariate probit model under the constraint defined in equation (12) and

the third panel shows the corresponding results under equation (13). The central conclu-

sions of this paper are, however, not sensitive to the treatment of fixed effects and we focus

our discussion on the case defined by equation (12) for the sake brevity. The estimated

magnitudes of correlations between unobserved determinants of parent and children’s non-

farm participation are similar: 0.21 for daughters and 0.25 for sons. The estimates show

that the intergenerational correlation between mother’s and daughter’s nonfarm partici-

pation is highly statistically significant (t-value=4.81). The estimated coefficient αm is

positive and large in magnitude (0.685) with a marginal effect of 0.146. In contrast, for

sons, the estimated αf = −0.143 with a t-value of 1.74. The results in panel 1 of Table 4

thus strengthen our central conclusion from the sensitivity analysis in section (4.1) above

that the estimated partial correlation in the nonfarm participation of mother and daughter

is not likely to be driven entirely by the genetic correlations; at least part of the occu-

pational linkage seems causal, probably reflecting cultural inheritance through role model

effects, learning externalities and transfer of reputation and social capital from mother to

the daughter as discussed in the conceptual framework above. In contrast, the lower bound

estimate for the correlation in father’s and son’s nonfarm participation is negative imply-

ing that the observed (positive) intergenerational correlation may be an artefact of genetic

transmissions across generations.35

To ensure robustness of our findings, we also check whether these results are driven

by the joint normality assumption underlying the bivariate probit model. Following AET

35We caution here that the fact that the lower bound estimate is negative for the father and son should notbe taken as conclusive evidence for an absence of intergenerational linkage. As mentioned before the lowerbound estimates are likely to underestimate the strength of occupational linkage given that the selectionon observables is likely to dominate. This also implies that the evidence from the bounds estimates infavor of a causal effect of mother’s non-farm participation is very strong.

22

(2005), we utilize the following semi-parametric specification for the error terms:

u = θ + u∗

ξ = θ + ξ∗

Where u∗ and ξ∗ are independent standard normals and θ is unrestricted. Bivariate

probit model estimated earlier is thus a special case where θ is assumed to be distributed

as normal. We estimated the model using nonparametric maximum likelihood method

suggested by Heckman and Singer (1984) and AET (2005). The estimation method treats

the distribution of θ as discrete; in practice, we obtain two points of support for θ. The

estimated ρ and αp are reported in the second and fourth panels of Table 4. Again, for the

sake of brevity, we focus on the case when the village fixed effects are treated as part of the

observables index (second panel in Table 4). The estimated magnitudes of ρ are smaller

compared with those from the bivariate probit model, but as before they are similar for

sons (0.178) and daughters (0.163). However, the overall results regarding the intergen-

erational effects remain unchanged. The effect of Nm (mother in nonfarm) on daughter’s

nonfarm participation is statistically highly significant, and positive (αm = 0.665). The

implied marginal effect is 0.135 which is virtually identical to that found in the bivariate

probit model (0.146). For sons, the estimated lower bound on intergenerational correla-

tion is positive but much smaller in magnitude (marginal effect=0.086) and statistically

insignificant.

Sources of Gender Differences

The main finding from our empirical analysis is that at least part of the partial cor-

relation in non-farm participation by mothers and daughters seems to survive when we

take into account the unobserved common factors across generations, but may not be so

in case of fathers and sons. Understanding the sources of this gender differences in the

intergenerational occupation linkages is clearly important. But a proper analysis of the

23

issue is beyond the scope of this paper. We, however, note a few plausible hypotheses

to be explored rigorously in future work. First, there is significant gender inequality in

educational attainment in Nepal in favor of sons (more than double). Education is an

important key to occupational mobility out of agriculture. The dramatically higher educa-

tional attainment of sons thus weakens the intergenerational link by enhancing their human

capital suitable for non-agricultural occupation. Education also broadens the set of role

models that include the teachers among others. Second, women in a traditional society

such as Nepal are brought up within the confines of a household with limited exposure to

the outside world. Naturally, they have their mother as the primary role model with a

disproportionately large influence on a daughter’s occupation choice. For sons, the set of

role models may extend well beyond the household; while the father may still exert some

influence, his impact on a son’s occupation choice becomes diluted. Third, as emphasized

in recent literature, occupational mobility may be linked to geographic mobility (see, for

example, Long and Ferrie (2007). Although geographic mobility is, in general, very re-

stricted in Nepal, there is clear gender effect in migration; the sons are, in general, more

likely to choose their geographic location as a strategy to move out of agriculture. The

geographic location of daughters is, in contrast, determined by the parental location (for

unmarried) or by the location of the husband (for married).36

(5) Conclusions:

The economic literature on intergenerational mobility has witnessed a renewed interest

in recent years. However, most of the existing economic research focuses on the income

correlations between father and son(s) in the context of developed countries. Using data

from a developing country, Nepal, we present evidence on the intergenerational occupational

mobility from agriculture to nonfarm sector with an emphasis on the gender differences.

It is extremely difficult, if not impossible, to find credible instrument(s) to address the

36We thank Bhashkar Mazumder for pointing out the possible role of geographic mobility. We, however,note that the role geographic location can play in our results is limited by the fact that we use village levelfixed effects. Since the village fixed effects are employed in separate daughters’ and sons’ regressions theymay not account for gender differences in locations completely.

24

genetic correlations (ability and preference) or other common unobserved determinants

of occupation choice. We employ the recent econometric approach developed by Altonji,

Elder and Taber (2005, 2000) to ascertain if the observed partial correlation in non-farm

participation can be attributed solely to genetic transmissions or at least part of the effect

is likely to be causal. The approach uses the degree of selection on observables as a guide to

the degree of selection on unobservables. It allows us to estimate lower bounds on the part of

the intergenerational occupational correlations that can be attributed to intergenerational

‘cultural inheritance’ possibly due to factors like role model effects, learning externalities,

and transfer of reputation and social capital, among other things. The results show that

the observed partial correlation between the father and a son can be easily explained

away by a moderate correlation in genetic endowments across generations. In contrast,

for the mother and daughter(s), the intergenerational occupational linkage is very strong,

and it is unlikely that the estimated partial correlation is driven solely by the unobserved

genetic correlations. The evidence points to a causal effect of mother’s occupation choice

on that of the daughter. The estimated lower bound for mother-daughter intergenerational

occupational correlation shows a marginal effect of 0.14. The evidence that, beyond the

genetic transmissions, there is persistence in nonfarm participation between the mother and

a daughter has important implications. It provides a link in the analysis of poverty trap

in rural households and brings into focus the intergenerational occupational linkage as an

important factor in understanding the restricted economic mobility of women in developing

countries.

References

(1) Altonji, J, T.E. Elder and C.R. Taber (2005): “Selection on Observed and Un-

observed Variables: Assessing the Effectiveness of Catholic Schools”, Journal of Political

Economy, 2005, Vol.113(1): 151-184.

(2) Altonji, J, T. Conley, T. Elder, and C. Taber (2005): “Methods for Using Selection

on Observed Variables to Address Selection on Unobserved variables”, Working Paper,

25

March, 2005.

(3) Altonji, J, T. Elder, and C. Taber (2000): “Selection on Observed and Unobserved

Variables: Assessing the Effectiveness of Catholic Schools”, NBER Working Paper No.

7831.

(4) Altonji, J.G. and T. Dunn (2000): An Intergenerational Model of Wages, Hours,

and Earnings, Journal of Human Resources, Vol. 35, n.2, 221-58.

(5) Arrow, K, S. Bowles, S. Durlauf (2000): Meritocracy and Economic Inequality,

Princeton University Press.

(6) Bardhan, P (2005): Theory or Empirics in Development Economics, Economic and

Political Weekly.

(7) Becker G. S. and N. Tomes (1979), “An Equilibrium Theory of the Distribution

of Income and Intergenerational Mobility,” Journal of Political Economy, vol. 87, no. 6,

1153-1189.

(8) Becker G. S. and N. Tomes (1986), “Human Capital and the Rise and Fall of

Families,” Journal of Labor Economics, vol.4, no.3, pt.2, S1-S39.

(9) Behrman, J. A. and M. Rosenzweig (2002): “Does Increasing Women’s Schooling

Raise the Schooling of the Next Generation?”, American Economic Review, March, 2002,

Vol. 92, no.1, pp. 323-34

(9) Behrman, J, A. Gaviria, M. Szekely (2001): Intergenerational Mobility in Latin

America, Working paper, IADB, Washington DC.

(10) Bjorklund, A, M Jantti, and G. Solon (2007): Nature and Nurture in the Inter-

generational Transmission of Socioeconomic Status: Evidence from Swedish Children and

Their Biological and Rearing Parents, NBER Working paper 12985.

(11) Bjorklund, A, M. Lindahl, and E. Plug (2006): The Origins of Intergenerational

Associations: Lessons from Swedish Adoption Data, Quarterly Journal of Economics, 121,

pp. 999-1028.

(12) Blanden, J, P. Gregg, S. Machin (2005): Intergenrational Mobility in Europe and

North America, Center For Economic Performance Report, April, 2005.

26

(13) Bowles, S, H. Gintis, M. Osborne Groves (2005): Unequal Chances: Family Back-

ground and Economic Success, Priceton University Press.

(14) Birdsall, N and C. Graham (eds) (1999): New markets, New Opportunities? Eco-

nomic and Social Mobility in a Changing World, Brookings Institution.

(15) Chadwick L. and G. Solon (2002), “Intergenerational Income Mobility among

Daughters,” American Economic Review, vol. 92, no. 1, 335-44.

(16) Cunha, F, J. Heckman and S. Navarro (2005): Counterfactual Analysis of Inequal-

ity and Social Mobility, in Fields, Gursky, amd Morgan ed. (2005).

(17) Dearden, L and S. Machin, and H. Reed (1997): Intergenerational Mobility in

Britain, Economic Journal, January, pp.47-66.

(18) Dunn, C (2004): The Intergenerational Transmission of Lifetime Earnings, Unpub-

lished Ph.D. dissertation, University of Michigan,

(19) Dunn, T. and D. Holtz-Eakin (2000), “Financial Capital, Human Capital, and the

Transition to Self-Employment: Evidence from Intergenerational Links,” Journal of Labor

Economics, Vol. 18, no.2, 282-305.

(20) Ferreira, F. H. G., and P. Lanjouw (2001), “Rural Nonfarm Activities and Poverty

in the Brazilian Northeast,” World Development, vol. 29, no. 3, 509-28.

(21) Fields, G (2001): Distribution and Development, MIT Press.

(22) Fields, G, D. Grusky, and S. Morgan (Eds) (2005): Mobility and Inequality: Fron-

tiers of Research from Sociology and Economics, Stanford University Press.

(23) Goldberger, A. (1989), “Economic and Mechanical Models of Intergenerational

Transmission,” American Economic Review, vol. 79, no.3, 504-513.

(24) Griffiths, A, W. Gelbart, J. Miller, and R. Lewontin (1999): Modern Genetic

Analysis, W. H. Freeman and Co., 1999.

(25) Hertz, T (2001): Education, Inequality and Economic Mobility in South Africa,Ph.D.

Dissertation, University of Massachusetts.

(26) Hertz, T (2005): Attrition and Age-Bias-Corrected Estimates of the Trend in the

Intergenerational Persistence of Family Income, Working paper, AU.

27

(27) Heckman J, and B. Singer (1984): A Method for Minimizing the Impact of Dis-

tributional Assumptions in Econometric Models for Duration Data, Econometrica, Vol.

52(2): 271-320.

(28) Lam, D, and R. F. Schoeni (1993), “Effects of Family Background on Earnings and

Returns to Schooling: Evidence from Brazil”, Journal of Political Economy, 101, 710-740.

(29) Lanjouw, P. and G. Feder (2001), “Rural Non-farm Activities and Rural Develop-

ment,” Rural Strategy Background Paper # 4, World Bank.

(30) Lentz, B.S., and D. N. Laband (1983), “Like Father, Like Son: Toward an Economic

Theory of Occupational Following.” Southern Economic Journal, vol. 50, no.2, 474-93.

(31) Lillard, L, and M. R. Kilburn (1995), “Intergenerational Earnings Links: Sons and

Daughters”, RAND, Labor and Population Program WP# 95-17.

(32) Long, J, and J. Ferrie (2007): The Path to Convergence: Intergenerational Occu-

pational Mobility in Britain and the US in Three Eras, Economic Journal, March 2007, pp.

C61-C71.

(33) Mazumder, B (2005): Fortunate Sons: New Estimates of Intergenerational Mobility

in U.S. Using Social Security Earnings Data, Review of Economics and Statistics, May,

2005.

(34) Mookherjee, D and D. Ray (2006): Occupational Diversity and Endogeneous In-

equality, Working Paper, Boston University and NYU.

(35) Morgan, S (2005): Past Themes and Future Prospects for Research on Economic

and Social Mobility, in Fields, Gursky, and Morgan (eds) (2005).

(36) Mulligan, C. B. (1999), “Galton vs. the Human Capital Approach to Inheritance,”

Journal of Political Economy, vol. 107 (6 Part 2), S184-S224.

(37) Plomin, R, DeFries, J, P. McGuffin, G. McClearn, and (2001): Behavioral Genetics,

Worth Publishers, New York, 4th Edition.

(38) Rosenbaum, P.R., 1995, Observational Studies, New York: Springler-Verlag.

(39) Rosenbaum, P.R., and D. B. Rubin, 1983, “Assessing Sensitivity to an Unobserved

Binary Covariate in an Observational Study with Binary Outcome,” Journal of Royal Sta-

28

tistical Society, Vol. 45(2): 212-18.

(40) Sjogren, A. (2000), “Occupational choice and Incentives: The Role of Family Back-

ground,” IUI, The Research Institute of Industrial Economics, Sweden, Working Paper #

539.

(41) Smith, T, C (1988), Native Sources of Japanese Industrialization, 1750-1920, Uni-

versity of California Press, Berkeley.

(42) Solon, G (2002): “Cross-Country Differences in Intergenerational Earnings Mobil-

ity”, Journal of Economic Perspectives, Summer 2002.

(43) Solon, G (1999), “Intergenerational Mobility in the Labor Market,” in O. Ashen-

felter and D. Card eds Handbook of Labor Economics, Vol 3A, Elsevier Science, North-

Holland, Amsterdam.

(44) Streufert, P. (2000), “The Effect of Underclass Social Isolation on Schooling Choice,”

Journal of Public Economic Theory, vol. 2, no.4, 461-82.

(45) Thomas, D. (1994), “Like Father, Like Son; Like Mother, Like Daughter: Parental

Resources and Child Height,” Journal of Human Resources, vol. 29, no. 4, 950-88.

(46) World Development Report (WDR) (2005): Equity and Development, The World

bank.

Table 1: Nonfarm participation of children conditionalon Parent's Employment Status (weighted mean)

Daughters SonsMother's employment inFarm 0.13 0.43Nonfarm 0.54 0.63Difference 0.414*** 0.20***Father's employment inFarm 0.14 0.41Nonfarm 0.24 0.57Difference 0.10*** 0.16***Unconditional 0.16 0.44

* significant at 1%; ** significant at 5%; *** significant at 1%

Probability of beingEmployed in Nonfarm Activities

Table 2: Probit Estimates of Intergenerational Correlations

(1) (2) (3)

Father in Non-Farm (nf) -0.072 -0.062 -0.068

[0.016] [0.012] [0.009](0.65) (0.49) (0.46)

Mother in Non-farm (nm) 1.286 1.283 1.157

[0.433] [0.406] [0.299](8.21)** (8.04)** (6.33)**

Pseudo-R2 0.1 0.522 0.622

Father in Non-Farm (nf) 0.372 0.402 0.260

[0.146] [0.157] [0.099](4.92)** (4.85)** (2.79)**

Mother in Non-farm (nm) 0.258 0.252 0.203

[0.102] [0.099] [0.078](2.16)* (2.00)* (1.38)

Pseudo-R2a 0.035 0.227 0.527

Individual and household characteristicsb No Yes YesVillage fixed effect No No YesNote.- Entries are probit coefficients. Standard errors are corrected for intra-cluster correlations due to clustered sampling. t-values are in parentheses and marginal effect of each variable (evaluated at sample means) is shown in bracket.

a. Pseudo R2 is defined as Var(X⁄γ)/[1+Var(X⁄γ)]b. Regressors in coulmn 2 include level of education, age, age squared, dummy for married, household size & composition, inherited land, distance to bank, un-earned income, dummy for migrant member in the household, 3 ethnicity dummies, father, mother and spouse's education level. Regressors in column (3), in addition to above regressors, include share of nonfarm employment by age cohort, and an index of village fixed effect.* significant at 5%; ** significant at 1%

Daughter's sample

Son's Sample

Table 3: Estimates of Intergenerational Correlations for different values of correlation of disturbances in bivariate probit models

Correlation

of

Disturbances Unrestricted Samplea Restricted Sampleb Unrestricted Samplea Restricted Sampleb

ρ=0 1.116 1.260 0.304 .323[0.284] [0.353] [0.117] [0.126](7.68)** (8.02)** (3.60)** (3.66)**

ρ=0.1 0.917 1.075 0.130 0.151 [0.216] [0.288] [0.049] [0.058] (6.35)** (6.88)** (1.54) (1.72)ρ=0.2 0.715 0.886 -0.046 -0.023 [0.154] [0.225] [-0.017] [-0.009]

(5.00)** (5.73)** (0.55) (0.26)ρ=0.3 0.509 0.692 -0.222 -0.198

[0.100] [0.165] [-0.082] [-0.075](3.63)** (4.56)** (2.72)** (2.32)*

ρ=0.4 0.299 0.492 -0.400 -0.374[0.053] [0.11] [-0.143] [-0.139](2.19)* (3.33)** (5.02)** (4.49)**

ρ=0.5 0.085 0.286 -0.578 -0.551[0.013] [0.06] [-0.201] [-0.20]

(0.64) (2.00)* (7.50)** (6.84)**N 2037 1126 2919 2547Note.- Entries are probit coefficients. Standard errors are corrected for intra-cluster correlations due to clustered sampling. t-values are in parentheses and marginal effect of each variable (evaluated at sample means) is shown in bracket.a.Regressors include level of education, age, age squared, dummy for married, household size & composition, inherited land, distance to bank, un-earned income, dummy for migrant member in the household, 3 ethnicity dummies, father, mother and spouse's education level, share of nonfarm employmentby age cohort, and an index of village fixed effect.b.Regressors include level of education, age, age squared, dummy for married, household size & composition, inherited land, distance to bank, un-earned income, dummy for migrant member in the household, 3 ethnicity dummies, father, mother and spouse's education level, share of nonfarm employmentby age cohort, and an index of village level dummies.* significant at 5%; ** significant at 1%

Daughter's Sample

Mother in Non-farm (nm)

Son's Sample

Father in Non-Farm (nf)

Table 4: Estimates of Lower Bounds for the Intergenerational Correlations

ρ αm ρ αf

ρ=Cov(Zγ2, Zβ2)/Var(Zγ2) 0.214 0.685 0.255 -0.143(1.70) [0.146] (6.71)** [0.053]

(4.81)** (1.74)

ρ=Cov(Zγ2, Zβ2)/Var(Zγ2) 0.163 0.665 0.178 0.086(0.24) [0.135] (8.09)** [0.03]

(3.50)** (0.04)

ρ=Cov(Xγ1, Xβ1)/Var(Xγ1) 0.219 0.677 0.257 -0.146(1.43) [0.144] (4.59)** [0.054]

(4.75)** (1.77)

ρ=Cov(Xγ1, Xβ1)/Var(Xγ1) 0.177 0.665 0.156 0.086(0.20) [0.135] (7.80)** [0.03]

(3.50)** (0.04)Note.- Entries are probit coefficients. Standard errors are corrected for intra-cluster correlations due to clustered sampling. t-values are in parentheses and marginal effect of each variable (evaluated at sample means) is shown in bracket.Regressors include level of education, age, age squared, dummy for married, household size & composition, inherited land, distance to bank, un-earned income, dummy for migrant member in the household, 3 ethnicity dummies, father, mother and spouse's education level, share of nonfarm employment by age cohort, and an index of village fixed effect.* significant at 5%; ** significant at 1%

Bivariate Probit Estimation

Nonparametric Maximum Likelihood Estimation

Nonparametric Maximum Likelihood Estimation

Daughter's Sample Son's Sample

Bivariate Probit Estimation

Table A.1: Summary Statistics by parental employment status.

Farm Non-farm Difference Farm Non-farm Difference(N=1880) (N=157) (N=2309) (N=610)

Participation in Nonfarm employment (proportion) 0.13 0.54 0.414*** 0.39 0.54 0.15***

Level of Education (Years) 1.63 2.66 1.03** 3.76 5.7 1.94***

Father's level of education (years) 1.15 2.01 0.96* 0.81 2.39 1.58***

Mother's level of education (years) 0.1 0.3 0.2 0.05 0.15 0.09*

Spouse' level of education (years) 2.07 2.05 -0.03 0.334 0.31 -0.024

Age 32.72 30.67 -2.05 37.13 28.88 -8.25***

Age squared 1230 1153 -77 1591 1020 -571***

Married 0.83 0.63 -0.19*** 0.8 0.63 -0.17***

Household size(log) 2.01 2 -0.001 1.77 1.85 0.08***

Share of adult female 0.25 0.26 0.012 0.23 0.22 -0.01

Share of children 0.17 0.17 0.0001 0.15 0.14 -0.01

Share of Young 0.35 0.35 0.006 0.34 0.38 0.04***

Share of Old 0.03 0.02 -0.01 0.02 0.02 0.004

Inherited land (value in million Rs.) (log) 9.26 6.65 -2.61*** 8.39 7.47 -0.92***

Travel time to nearest bank 2.65 1.78 -0.87*** 2.91 2.34 -0.57***

Un-earned income (million Rs) 0.008 0.007 -0.001 0.005 0.005 0.0008

Migrant in the household 0.39 0.48 0.09 0.32 0.46 0.14***

Upper caste Hindu (Proportion) 0.36 0.21 -0.15** 0.36 0.28 -0.08***

Lowr caste Hindu (Proportion) 0.07 0.19 0.12* 0.07 0.15 0.08***

Tribal (Proportion) 0.29 0.23 -0.06 0.26 0.26 -0.006

Share of nonfarm in district by age cohort 0.14 0.2 0.06*** 0.28 0.3 0.02***

* significant at 1%; ** significant at 5%; *** significant at 1%

Mother's employment inDaughter's Sample Son's Sample

Father's employment in

Appendix Table A.2: Probit Estimation of Intergenerational correlations

Daughter in Son inNon-Farm Sector Non-Farm Sector

Mother in Non-farm Emplyoment 1.283 0.252(8.04)** (2.00)*

Father in Non-farm Employment -0.062 0.402(0.49) (4.85)**

Level of education (year) 0.070 0.028(5.05)** (3.80)**

Age 0.055 0.107(2.29)* (7.16)**

Age Squared -0.001 -0.001(1.85) (7.28)**

Married (Yes=1) -0.075 0.375(0.56) (4.10)**

Household Size (log) -0.293 -0.291(3.26)** (3.70)**

Share of Adult Female -0.150 -0.385(0.28) (0.96)

Share of Children 0.337 0.385(0.84) (1.22)

Share of Youth 0.452 -0.193(1.21) (0.76)

Share of Old 0.118 -0.423(0.20) (0.81)

Inherited Land (log) -0.048 -0.008(5.84)** (1.28)

Father's education (year) -0.026 0.0003(1.56) (0.03)

Mother's Education (year) 0.034 0.098(0.97) (2.18)*

Spouse's education(year) 0.031 0.017(2.92)** (0.92)

Travel time to Bank 0.007 0.008(0.63) (1.04)

Unearned income -12.758 0.038(3.29)** (0.12)

Migrant in the household (yes=1) -0.005 0.504(0.06) (8.25)**

Upper Caste Hindu (yes=1) 0.089 -0.322(0.81) (4.46)**

Lower Caste Hindu (yes=1) 0.100 0.227(0.67) (2.00)*

Belongs to tribe (yes=1) 0.290 -0.170(2.63)** (2.11)*

Constant -1.581 -1.887(3.25)** (5.41)**

Observations 2037 2919Note.- Entries are probit coefficients. Standard errors are corrected for intra-cluster correlations due to clustered sampling. t-values are in parentheses.* significant at 5%; ** significant at 1%

Employment Status